Functional summarization of gene product clusters using Gene Ontology similarity measures

The paper addresses the problem of constructing a functional summarization of groups of gene products that are found by clustering a database of such products annotated by the Gene Ontology. Our method builds the "most representative term" (MRT) for each cluster in three increasingly sensitive ways. Initially, we perform crisp hierarchical clustering using BLAST and our novel fuzzy measure similarities and find the MRTs as the terms of highest frequency in the description of the gene products. Using weights from the fuzzy partition matrix generated by a relational fuzzy clustering algorithm, we show how more specific MRTs can be made. Finally, weighting these memberships by the information content of each term further increases the specificity of the functional annotation of the clusters.